Educational inequalities in mortality and their mediators among generations across four decades: nationwide, population based, prospective cohort study based on the ChinaHEART project

Study design and population

The ChinaHEART project is a government funded public health programme designed for screening of cardiovascular disease risk and for intervention in community based populations throughout China. The design of ChinaHEART project has been described previously.12 Briefly, from 20 November 2014 through 31 December 2021, 351 county level regions in all 31 provinces in the mainland of China were selected as study sites to provide diversity in geographical distribution, population structure, and exposure to risk factors and disease patterns (supplementary method). Study site selection also considered population size, population stability, and local capacity to support the project. A total of 12 323 531 residents aged 35-75 years, who had lived in the community for at least six months in the previous 12 months, were invited in this project, and about 4.5 million residents were recruited.

In baseline interviews of the 4.5 million ChinaHEART participants, people with the serial project ID number ending with 1, 3, 5, or 7, were selected as a representative sample to provide more detailed information of behavioural risk factors and morbidities at the baseline. Among them, 1 295 871 participants born between 1940s and 1970s were considered eligible for this study. After exclusion of 12 097 (0.93%) participants with incomplete data for education status data (n=11 886) or date of death (n=211), 1 283 774 remained for analysis. Missing data were noted in 8.8% of participants and were mainly related to lifestyle. We used a widely accepted random forest method to conduct the imputation for categorical and the continuous variables1314 (missForest package in R; supplementary method).

Data collection and variable definition

For each participant, a standardised interview was conducted in person by trained personnel to collect information on education and other socioeconomic status (ie, annual household income, health insurance, marital status, and occupation), lifestyle (ie, tobacco smoking, alcohol drinking, diet patterns, and leisure time physical activity; supplementary method), and medical history at baseline. Education was categorised according to the International Standard Classification of Education scale: primary school or below was equivalent to level 0 and 1; middle school was equivalent to level 2; high school was equivalent to level 3 and 4; and college or above education was equivalent to level 5 to 8.15

For behavioural factors, tobacco smoking status was categorised into never, formerly, or currently smoking. Alcohol drinking frequency was categorised into never, once or less per month, two to four times per month, two to three times per week, or more than four times per week. The amount of alcohol consumption during a typical drinking day was also estimated. Leisure time physical activity was quantified by typical types of activity at different intensity (swimming, running, or aerobic exercise were vigorous intensity activities; ball games, walking, gymnastics, folk dancing, Tai-Chi, qigong, or other exercise were moderate intensity activities), frequency, and exercise duration per week. A qualitative food frequency questionnaire was used to collect habitual dietary intake by asking frequency of eating 12 food components over the previous year (supplementary method). For each food component, five frequency categories were provided to choose (ie, daily, four to six days per week, one to three days per week, one to three days per month, and never or almost never).

More specifically, a participant with healthy behaviours was defined as: never smoking or formerly smoking who stopped for reasons other than illness16; never drinkers, or drinkers who drank no more than 25 g (three units) for men or 15 g (two units) for women per day on average; individuals having sufficient leisure time physical activity (ie, at least 150 min moderate intensity aerobic activities or 75 min vigorous intensity aerobic activities per week); and individuals with a healthy diet score of 4 or higher based on weekly food intake of six components (ie, daily intake of fresh fruit, fresh vegetables, and whole grains; eating fish and other seafood at least one day per week, bean and bean food at least four days per week, and red meat fewer than seven days per week).

As metabolic factors, blood pressure, blood glucose, lipids, weight, height, and waist circumference at baseline were measured using the standardised protocols and unified devices (supplementary method). Hypertension was defined as systolic blood pressure of 140 mmHg or more, diastolic blood pressure was 90 mmHg or higher, or a self-reported disease history of hypertension, or the use of anti-hypertensive medications, according to the US and Chinese definitions.171819 Diabetes was defined as blood glucose greater than 7 mmol/L after at least 8 h after last meal or random blood glucose greater than 11.1 mmol/L according to the Chinese Guidelines for prevention and treatment of type 2 diabetes, or a self-reported disease history of diabetes, or usage of hypoglycaemic agents. Dyslipidaemia was defined as an elevated total cholesterol (≥6.2 mmol/L) or low density lipoprotein (≥4.1 mmol/L) according to guidelines for dyslipidaemia in Chinese adults,20 or self-reported lipid lowering drugs usage. Body mass index was defined as weight in kilograms divided by the square of height in metres. History of cardiovascular disease was defined as self-reported medical histories or received the treatments, including myocardial infarction, coronary artery bypass graft surgery, percutaneous coronary intervention, and stroke.

Ascertainment of outcomes

Participants’ vital status and cause of death were collected through a passive follow-up process in which a linkage of data was established between the cohort and the National Mortality Surveillance System and Vital Registration of China Center for Disease Control and Prevention that covers urban and rural areas in all the 31 provinces of the mainland of China. The death records in this system were reported by health care institutions almost in real time then checked against local residential records and health insurance records annually.

In this National Mortality Surveillance System and Vital Registration, the main cause of death was coded using the 10th edition of the International Classification of Diseases (ICD-10). We considered that cardiovascular disease is the leading cause of death in China, and behavioural and metabolic factors are the leading risk factors for cardiovascular disease incidence and prognosis. Cardiovascular disease mortality (ICD-10 codes I00-I99) was also included as a study outcome in our analyses.

Statistical analyses

We described participant characteristics by education levels using percentages for categorical variables and mean (standard deviation) or median (interquartile range) as appropriate for continuous variables. To compare the trends of characteristics from higher to lower education levels, analysis of variance tests for continuous variables and Mantel-Haenszel χ² test for categorical variables were conducted.

To evaluate the cumulative influences of socioeconomic, behavioural, and metabolic factors on the educational inequalities, a series of stepwise Cox proportional hazard models were used to calculate hazard ratios and 95% confidence interval for the education level with all cause mortality in different generations and in overall cohort. Socioeconomic, behavioural, and metabolic factors were adjusted in sequence. In model 1, we adjusted for age, sex, urbanity, and geographical region. In model 2, socioeconomic factors (ie, marriage status, household income, medical insurance, and whether work as a farmer) were additionally adjusted for. In model 3, healthy lifestyle behaviours (ie, never smoking or stopped smoking for reasons other than illness, drinking no alcohol or no more than 25 g (for women) or 15 g (for men) per day on average, having sufficient leisure time physical activity, and having a healthy diet score ≥4) were added too. In model 4, metabolic factors (ie, total cholesterol, blood glucose, systolic blood pressure, diastolic blood pressure, body mass index, and waist circumference) were further included. To evaluate the separate influence of socioeconomic, behavioural, and metabolic factors on the educational inequalities, Cox proportional hazard models were fitted in adjustment to socioeconomic, behavioural, and metabolic factors separately. We tested the proportion hazard and linear assumption of the Cox models (supplementary method). Considering that the relation between these metabolic factors and outcomes were not linear, we divided each of them into 20 groups and then included them as dummy variables into the model. As the mortality data were available up to 31 December 2021, we censored follow-up at this date or the date of death, whichever occurred first in all models.

The Kunst-Machkenbach’s relative index of inequality was calculated by a structure regression framework to estimate the extent of educational inequality in mortality and how this inequality varied among people from different generations.21 Relative index of inequality can be interpreted as the ratio of the mortality rate predicted for the hypothetical lowest end of the education continuum to the rate predicted for the hypothetical highest end. To evaluate the effects of three categories of potential contributing factors (ie, behavioural, metabolic factors, and socioeconomic factors) on the educational inequalities in mortality, relative index of inequalities were estimated by fitting Cox regression models adjusted for each category, in addition to age and sex. According to the causal structure of the association between education and mortality (supplementary method), mediation analyses were conducted using the public %MEDIATE SAS macro (https://ysph.yale.edu/cmips/research/software/mediate) to estimate the mediating effects of those factors on the relation between education levels and all cause mortality among people from different generations. Mediators’ effects were reported, if existed, as the proportion of the relationship mediated through them.

We conducted several sensitivity analyses. Firstly, we repeated all analyses in participants without missing value in behaviours or socioeconomic factors (n=1 186 327). Secondly, we conducted causal mediation analysis using the mediation R package to compare the results from the traditional mediation analysis. Thirdly, as a result of the significantly different distributions of education levels across sex and urbanity, we conducted all analyses stratified by sex (female and male) and urbanity (rural and urban). Fourthly, to further assess the health effect of education inequality on cardiovascular diseases, we repeated all analyses with cardiovascular disease mortality as the outcome. Lastly, we used polytomous variables of health lifestyle behaviours in Cox regression models to test the influence of binary variables.

A two side P value of <0.05 was considered statistically significant. All analyses were conducted with SAS 9.4 (SAS Institute Inc, Cary, NC, USA) and R 4.1.1.

Patient and public involvement

No participants were involved in setting the research question or the outcome measures, nor were they involved in developing plans for the design or implementation of the study. No participants were asked to advise on interpretation or writing up of the results. We know that the public involvement has great value in improving the quality of research; however, we did not have the necessary funding to invite participants for their advice. We intend to engage participants and the public in the dissemination of the results of our study.